How your corporation can profit from plentiful computing energy within the age of AI

The high-frequency trading firm Citadel Securities, for example, began embedding advanced computing capabilities into its core strategy in the early 2000s, and in turn was able to leverage immense processing power and technologies, such as machine learning, to execute trades in microseconds. Citadel’s ability to rapidly put large amounts of computing power to use have made it the largest retail market-making firm in the U.S., executing over $400 billion in trades per day.

The innovations made possible by unconstrained computing power have only expanded with the development of AI. Yet for many, there appears to be a paradox looming, based on the projection that because frontier AI models are so computationally intensive to train and use, their proliferation could lead to a scarcity of computing power. 

To test the likelihood of this scenario, the BCG Henderson Institute and Exponential View teamed up to quantitatively model future supply of and demand for computing power. Our model incorporated moderate assumptions about future supply, while using bullish demand estimates. What we discovered was that the abundance of computing power is likely to continue even in an aggressive scenario of rapid generative AI adoption.

While tech CEOs have gotten this message and have geared up to leverage the continued availability of affordable computing power, many business leaders have not fully digested the potential reach of this reality in conjunction with the transformative possibilities of AI—even though it could have a significant impact on how companies of all stripes organize and operate. Viewing change as incremental, rather than exponential, is a common mistake that Nathan Myhrvold, Microsoft’s first-ever chief technology officer, observed as far back as 1993, writing at the time that most executives “act like linear extrapolators that won’t know what hit them.”

For executives seeking to avoid being blindsided by an exponential wave of change brought on by teaming AI with expansive computing power, now is the time to reflect on the implications. In what novel ways might leaders create, deliver, and capture value enabled by orders-of-magnitude more computing power? Having an answer to that question as AI and abundant compute power continue to feed off each other will be key to staying at the forefront of innovation and competition. 

Computing power will continue to abound

How likely is computing power to become scarce in the age of ever more computationally intensive generative models? Will the world literally run out of the high-end semiconductors needed to support GenAI workloads? To examine this question, our quantitative model estimated global computing power incorporating a bullish demand scenario for GenAI inference, where models keep getting larger over time and enterprise adoption of the technology is very aggressive by historical standards. (We focus our model on inference because in a world of widespread GenAI adoption, the compute used to train a model will be dwarfed over time by the aggregate compute spent using the models.) 

What we find is that even in the bullish demand scenario, aggregate computing power consumed for GenAI inference will still take up just a third of the expected overall computing power supply by 2028. In other words, we believe that the regime of widely available and affordable computing power that has propelled the digital economy will continue to hold even as GenAI adoption accelerates.

This is not to say that there aren’t real challenges for a sustained expansion of computing power supply. Hardware availability could be disrupted by geopolitical tensions impacting the semiconductor supply chain, while energy supply may constrain the expansion of data center capacity in some geographies. But advances in energy efficiency (through, e.g., new, specialized hardware), as well as efforts to make data centers self-sustaining through on-site energy generation, should help meet this energy challenge. 

Even with these risks, the overarching trend points toward abundant computing resources in the coming years with the power to fuel increasingly potent innovation as AI capabilities continue to expand and deepen.

How businesses can harness computing power and advanced AI

There is little question that computing power can be a source of competitive advantage for businesses that are quick to harness it. We already mentioned the case of Citadel in financial services, but examples abound in other industries, too. Amadeus, for example, originally a traditional European travel booking system, adopted open source systems and cloud computing early on, allowing it to process more than 100,000 transactions per second, and transformed into a global travel technology leader supporting airlines and travel agencies worldwide. 

AI makes computing power more potent still, as it can be used to accelerate or even automate increasingly complex problems. As we have argued in this column before, AI promises to render much of current human cognitive activity in the workplace tractable for machines, in the same way that the old factory floors now are often occupied by robots, rather than human workers. Moderna, for example, has been able to disrupt the pharmaceutical industry, accelerating vaccine development from five-plus years to just months—all thanks to the compute- and AI-powered automation of complex scientific processes.

These cases illustrate how much can be gained by being prepared to deploy computing power, especially in the age of AI. CEOs would be wise not to dismiss the impact of computing power for creating or maintaining a competitive edge—lest they run the risk of being disrupted by forward-thinking competitors.

From limitations to possibilities: How imagination drives the next wave of innovation

As computing power continues to expand alongside the relentless advance of AI, it is imagination that will unlock new possibilities from AI and other advanced technologies. Consider other historical developments, such as the shift from dial-up/ISDN to DSL, then cable, fiber-optic, and now to satellite-based internet (e.g., Starlink), enabling global connectivity. In many cases, the tech capabilities for such a leap already exist, but future advancements can amplify them, spurring innovation beyond current expectations.

In anticipation of order-of-magnitude more compute in the near future, business leaders need to ask themselves what are the highest-complexity, highest-value processes and activities that have proven resistant to automation due to their computational complexity or cost. As our colleagues at Exponential View put it, “What would you do with 1000x more computing power? How would your organization use it? If you were to ask these questions to Sam Altman or Satya Nadella or Sundar Pichai, they would have an answer. Do you?”

The process of formulating an answer to this challenge can be more systematic than many leaders realize. First, it requires a deep examination of a company’s business model, starting with its value proposition. Compute is the key to reaching new heights in personalization and tailoring of goods and services, en route to a true “customer segment of one.” 

Next, leaders must take a hard look at their value creation and delivery. What are seemingly outlandish possibilities that become more plausible with sufficient compute? Picture an insurance company capable of developing a digital twin of the Earth to simulate and better forecast atmospheric and geological risks. This is a technological feat that is no longer out of reach with abundant computing power; it could transform an insurer’s approach to risk management, upending the current state of insurance market competition based on marginal improvements to traditional actuarial methods. 

Finally, when it comes to value capture, the combination of strong models and high-frequency, high-quality data creates an extraordinary opportunity to optimize pricing models. 

Of course, the (by now) old mantra of getting your data in order still applies. But the horizon of expanding computing power harnessed through AI is primarily one of imagination. And, again, there is a way to systematically turn complex organizations into “imagination machines.” As Martin Reeves and Jack Fuller have argued in their book, The Imagination Machine, business leaders can and should actively seek out surprises, in order to rethink their own mental models, and then put them to test in collision with the real world. If imagination is the ability to create a mental model of something that doesn’t exist yet, then nothing will be more important than imagination itself to success in an economy increasingly driven by the reality-expanding power of compute and AI.

***